[bugfix] fix ray start failed: local_world_size cannot little than visible device count error (#4457)

### What this PR does / why we need it?
Fix the ray start failed bug: local_world_size cannot little than
visible device count error
detail see issue #4456.

This fix code is copied from vllm fixing modify, PR:
[#28873](https://github.com/vllm-project/vllm/pull/28873)


- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2

---------

Signed-off-by: leo-pony <nengjunma@outlook.com>

tests/ut/torchair/test_torchair_worker.py

@@ -59,6 +59,7 @@ class TestNPUTorchairWorker(TestBase):
             worker.vllm_config = MagicMock()
             worker.parallel_config = MagicMock()
             worker.parallel_config.local_world_size = 0
+            worker.parallel_config.data_parallel_size = 1
 
             result = worker._init_device()
 
@@ -93,6 +94,7 @@ class TestNPUTorchairWorker(TestBase):
             worker.vllm_config = MagicMock()
             worker.parallel_config = MagicMock()
             worker.parallel_config.local_world_size = 0
+            worker.parallel_config.data_parallel_size = 1
 
             result = worker._init_device()
 

tests/ut/worker/test_worker_v1.py

@@ -329,6 +329,8 @@ class TestNPUWorker(TestBase):
             worker.model_config = MagicMock()
             worker.parallel_config = MagicMock()
             worker.parallel_config.local_world_size = 0
+            worker.parallel_config.data_parallel_size = 1
+
             worker.model_config.seed = 42
 
             # Test _init_device

vllm_ascend/worker/worker_v1.py

@@ -208,12 +208,18 @@ class NPUWorker(WorkerBase):
         NPUPlatform.set_device(device)
         NPUPlatform.empty_cache()
 
-        visible_device_count = (torch.npu.device_count()
+        if (self.parallel_config.data_parallel_size > 1
-                                if torch.npu.is_available() else 0)
+                and self.parallel_config.data_parallel_size_local > 0
-        assert self.parallel_config.local_world_size <= visible_device_count, (
+                and self.parallel_config.distributed_executor_backend
-            f"local_world_size ({self.parallel_config.local_world_size}) must be "
+                not in ["ray", "external_launcher"] and
-            f"less than or equal to the number of visible devices "
+                self.vllm_config.parallel_config.data_parallel_backend != "ray"
-            f"({visible_device_count}).")
+                and self.vllm_config.parallel_config.nnodes_within_dp == 1):
+            visible_device_count = (torch.npu.device_count()
+                                    if torch.npu.is_available() else 0)
+            assert self.parallel_config.local_world_size <= visible_device_count, (
+                f"local_world_size ({self.parallel_config.local_world_size}) must "
+                f"be less than or equal to the number of visible devices "
+                f"({visible_device_count}).")
 
         self.init_npu_memory = NPUPlatform.mem_get_info()[0]
         # Initialize the distributed environment.